Decoding Airline Passenger Sentiment Dynamics Using Transformer-Based Models and Time-Series Clustering

Abstract

This study proposes a novel framework, namely the Sentiment Trend Analysis on Reviews (STAR). The STAR framework incorporates time-series clustering into sentiment detection to discover distinct airline passenger sentiment patterns, followed by transformer-based analysis to identify customer concerns from review text that drives negative sentiments. By applying the STAR framework to U.S. airlines, we identified notable differences in sentiment trends between low-cost carriers (LCCs) and full-service carriers (FSCs). Passengers are consistently dissatisfied with LCCs' baggage handling procedures and fees throughout the year and are more concerned about bookings and refunds during the Christmas holidays. For FSC passengers, flight cancellations and delays rank among their top concerns even during periods without major holidays. This study provides guidance for researchers and airlines to respond to the dynamics of customers' expectations and design recovery actions.